As light from an emitting source, e.g., the Sun, is reflected by a geologic surface, its spectrum becomes the carrier of useful geologic information. In particular, the ratio of reflected-to-emitted light spectral fluxes, a quantity broadly defined as reflectance, is a complex convolution of how light interacted with the “atmosphere” as it travelled to and away from the geologic surface, of how it interacted with individual mineral crystals or grains on the surface, and of illumination geometry. In particular, Visible-Shortwave Infrared (VSWIR) reflectance is sensitive to mineral compositions and grain sizes in the geologic target. Thus, pending our ability to deconvolve these different signals, VSWIR spectroscopy is a prime tool to investigate planetary surfaces from orbit. However, the combined effects of mineral abundances, grain sizes, noise, and the non-linearity of radiative transfer models lead to an ill-posed inverse problem; in particular, several equally-good solutions may fit the data. Thus, a significant knowledge gap in VSWIR spectroscopy is that of the errors and uncertainties associated to the inverse determination of mineral abundances and grain sizes, and constraining those would represent a major improvement to commonly used inversion technique by shedding light onto the reliability of inferred compositions of planetary regoliths. To bridge this gap, we developed a probabilistic framework to remote compositional analysis of planetary surfaces. Although currently implemented for VSWIR wavelengths, the algorithm may be easily modified to use spectra acquired over different wavelength ranges.